Rosetta 3.5
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
Author
Steven Lewis smlew.nosp@m.i@gm.nosp@m.ail.c.nosp@m.om

Metadata

Code and documentation by Steven Lewis smlew.nosp@m.i@gm.nosp@m.ail.c.nosp@m.om. This document was last updated 6/24/11 by Steven Lewis. The PI was Brian Kuhlman, bkuhl.nosp@m.man@.nosp@m.email.nosp@m..unc.nosp@m..edu.

Code and Demo

The code for this application lives in rosetta/rosetta_source/src/apps/public/interface_design/anchored_design/AnchorFinder.cc. Its mover is embedded in the application file. There is a demo in the integration tests (rosetta/rosetta_tests/integration/tests/AnchorFinder). There is a more extensive demo with more documentation at rosetta/rosetta_demos/AnchoredDesign, or in the demo section of the release.

References

New_Library contains code for this purpose

The purpose of AnchorFinder is to search for anchors in protein dimers. It is part of the overall AnchoredDesign suite. Briefly, anchors are regions in an interface that serve as handles and could be grafted from a natural interacting partner into a new scaffold (the purpose of the AnchoredPDBCreator application) and then flexibly designed around to create a new protein-protein interface between the original target and scaffold (the AnchoredDesign application).

It was written to help look for proteins useable for fixed-sequence benchmarking of the AnchoredDesign protocol. It can also be used to examine individual interfaces for things that look like anchors (although the human eye is superior if you only want to look at a few structures).

Compiling the Rosetta library and application using SCons

AnchorFinder is intended to be run "robustly". See also the "Robust Rosetta" documentation, (hopefully this is a clickable link to Robust Rosetta). AnchorFinder was run against the entire PDB at one point; robustification allows it to not crash on malformed PDBs.

Algorithm

The idea is pretty simple. In the first phase, it throws out inputs that are not protein multimers. Nonprotein atoms are stripped, and monomers and too-short structures (20 residues) are thrown out.

If a candidate structure passes that filter, then AnchorFinder basically prints out the neighbor graph, sorted by chain, to a file called <pdbname>.out.

It then looks for anchor regions in that structure (regions of high loopness and many cross-interface neighbors) and collects those windows in another file, default name goodfile.out.

Limitations This code looks for anchors only in a vague sense, you still need to look at the suggested anchors before use. Also, it cannot tell if your input structures are biological heterodimers, homodimers, crystal dimers, octamers, 3 identical heterodimers in one asymmetric unit, etc. If you already know what protein you want to look for anchors in, your eyes are better than this code anyway: load it up in PyMOL and just look. If you have 20 possibilities it's useful.

Modes Nominally, "looking for benchmark proteins" is a mode (that only the original author will ever use), and "looking for anchors in protein dimer X" is a mode that you might use. The code is used identically, although you'd use a large number of inputs for the first and few for the second.

Input Files

This code has no special inputs, just the flags and pdbs.

Options

AnchorFinder options consist of flags to control what it looks for in an anchor and where it puts the data:

Suggested general rosetta options include:

Tips

The code is much more efficient if you use all the "general rosetta" options above.

It is slower in execution, but far faster because it doesn't crash, if you do the robustness tricks linked to above.

If the code accumulates large amounts of memory, you don't have a memory leak. What is happening is that the Job Distributor is needlessly caching the huge number of input PDBs, not knowing they are unneeded after first use. When this code was written, that was a fixed problem; it may have since become unfixed. In that case, run in smaller batches and concatenate the outputs.

Expected Outputs

Pdbname.data looks like this:

Rows are residues, columns are chains, data are neighbors in that chain for each residue residue chain PDBdata DSSP 1 2 1 1 2 D L 7 0 2 1 3 D L 10 0 3 1 4 D L 14 0 ...

You can mine this later without rerunning AnchorFinder to look for anchors. The number of columns will grow if you have more than two chains in the PDB (which can still be valid input, if you have an asymmetric unit of 3 heterodimers).

Goodfile.out looks like this:

PDB pdb2vk1 window 45 loopness 5 nbrs 0 28 0 0 start 46 A pymol select pdb2vk1 and chain A and resi 46-50 PDB pdb2vk1 window 108 loopness 5 nbrs 0 25 0 0 start 109 A pymol select pdb2vk1 and chain A and resi 109-113 PDB pdb2vk1 window 109 loopness 5 nbrs 0 36 0 0 start 110 A pymol select pdb2vk1 and chain A and resi 110-114 PDB pdb2vk1 window 110 loopness 5 nbrs 0 46 0 0 start 111 A pymol select pdb2vk1 and chain A and resi 111-115 PDB pdb2vk1 window 111 loopness 5 nbrs 0 46 0 0 start 112 A pymol select pdb2vk1 and chain A and resi 112-116 PDB pdb2vk1 window 112 loopness 5 nbrs 0 47 0 0 start 113 A pymol select pdb2vk1 and chain A and resi 113-117

Each line identifies the PDB, the window number, its loopness, its number of neighbors on each chain in the PDB (variable # of columns), the starting residue PDB numbering for the window, and a Pymol selection for the window.

Post Processing

At this point, the data is yours to play with. I searched for windows with large numbers of neighbors on only one chain using sifter.py (included in the protocol capture attached to publication), then sorted for those with the largest number of neighbors (sort -n -k1 <input>). After that it was all manual filtering to choose structures for the benchmarks. I've never used it to look for anchors in small numbers of structures.

New things since last release

Rosetta 3.3 is the first release.